A Refreshing Perspective on AI and Truth

Source: Dev.to

The Nature of Truth and AI

Everyone has a favorite movie. Some of us ask why. None of them are wrong; each is right relative to where they stand: their experience, their era, the conversations they’ve been part of. Truth, for humans, has an address.

How LLMs Process Information

During training, a model ingests millions of documents simultaneously—texts from opposing centuries, conflicting political movements, irreconcilable cultures—and flattens them into a single mathematical space. To a film historian, that 1921 Keaton film explains the 2026 blockbuster. To an AI, both exist at the same depth, in the same timeless fog. There is no “before.” There is no provenance.

When you ask an AI to review your article and it loves a sentence, then in the next session calls that same sentence weak, that isn’t a bug or a bad day. There is no plot, and there is no twist, because there is no story being told from anywhere.

When forced to answer, the model doesn’t reason from a position. It calculates a statistical average—blending the kid, the cinephile, and the historian into something that sounds authoritative because it contains all of them and is anchored by none of them.

The Role of Prompting

This is the core paradox: an LLM is never wrong because it is incapable of being right—not in the way that matters. Being right requires standing somewhere.

Which is why a good prompt is more important than most people think. The prompt is the only provenance the model has. It’s the only “when,” “who,” and “from where” available to it. A vague prompt doesn’t just get a vague answer—it gets an answer from nowhere, averaged from everywhere. A specific, contextual prompt is the closest thing an LLM has to a position in time.

Conclusion

So maybe “truth‑seeking AI” isn’t entirely a broken idea. It’s just that the seeking starts with—and depends on—you (whatever “you” really means).